Optimal...not sure this can be answered in the arbitrary percentage sense, but perhaps instead in the abstract sense.
Suppose you had 90% locked up in MNs...that would mean 10% was used for "regular commerce", or whatever we want to call it. Advantage of this is like it is with the gold market...it creates some relative level of stability to the price because the gold bugs don't sell at every trading indicator. Reverse that to 90% regular commerce and 10% in MNs, we probably get much wilder price swings, which discourages keeping it locked in MNs to some degree (compared to the flipped percentage in the first example). So, there is likely some tipping point for everyone but true believers in the tech ("Dash bugs") where if not enough is locked in MNs as a percent of the total, the price will fluctuate too wildly, discouraging MN counts from rising unless the number of Dash needed for a MN is reduced, or regular commerce recedes (not a good thing) to come into some equilibrium again with the MN share of the available DASH.
So, we can deduce that there is some moving target based upon several variables, but not some set percentage. Instead, the "optimal" percentage at any moment is the percentage which allows for MN runners who aren't Dash bugs (ride or die) to feel comfortable enough in relative price stability to keep their investment in place, or to add new MNs to that investment (or, alternatively, invest in their first MN due to relative price stability). The other side of this optimal percentage is enough Dash in circulation for regular commerce to keep liquidity high and allow for steady expansion of the user base, but not so low as to mess with liquidity and expansion, but also not so high as to cause relative price instability.
Of course, there are other factors which could change this view of "optimal". For example, if Dash were to implement some Keynesian (oh no!) mechanism like some central bank, where it used some amount of Dash to shift between MNs and Dash available for regular commerce, which would be like a central bank manipulating interest rates and money supply to cause artificial (and unsustainable!) stability, then obviously what it optimal and how to achieve it would change to a degree (as the intervention causes its own unintended consequences - see the failure to peg to the dollar in other crypto projects as a "sort of" example of this failed concept).
So, my answer is: The "optimal" percentage of Dash in MNs versus regular commerce is a moving target that differs from moment to moment, but based upon what I know now about Dash, and what variables I can therefore consider, the optimal percentage is X, where X is neither 0% nor 100%, but some non-zero percentage for both MNs and regular commerce. These two optimal percentages (MNs and regular commerce) are likely to be weighted more toward MNs than regular commerce, based upon how relative price stability is achieved (short and medium term) in present fiat currencies and commodities, like gold. The important considerations are a high enough MNs percentage to create relative price stability and maintain MN investment and facilitate steady growth proportional to user growth, and yet high enough percentage for regular commerce to keep high enough availability to maintain liquidity relative to user growth.
I'd venture a guess that as the project stands, 60% isn't a bad number, and I wouldn't worry at even 70% or 80%. As the user base grows, and the interest bearing accounts in Evolution come into being, the optimal percentage will likely drop, as parking your Dash, like say 100 Dash, will earn you interest like current MNs, so that will perpetuate stability in price naturally, even without the consideration they are indeed a partial MN, for all intents and purposes, at that point. Gold, for example, saw a bottom for the market at about $1050 per troy oz. It has bounced off that bottom and is now at about $1300 an ounce. It's gone to almost $1400, and hasn't (yet) dipped below $1300 since reaching that mark some months ago. At $1050, it was mostly governments and gold bugs holding gold, as it is not used as a form of money much these days (digital fiat or paper/coin fiat are mostly used worldwide). You can relate this to MNs and their holdings. This created price stability, relatively, for gold (hence, the bottom was found and it hasn't approached it since). As more people began to invest in gold after the bottom was in, for panic reasons or for inflation hedging to some degree, the price began to rise, but so did the volatility (as there was a lot more daily trading going on). This added trading and relative price fluctuations can be related to Dash for "regular commerce". What would be optimal, so to speak, for gold, is the same as for Dash, in theory...you want long term holders of gold to reach an equilibrium with short term traders of it to create A) relative price stability and available liquidity no matter how many people use it or not, which will inspire further trust in it and therefore further growth of the market. In other words, the "bugs" and the "users" are locked in a sort of ever-changing balancing act, or dance, and the "optimal" percentage for each is based upon user volume and demand, growth or recession rate in this regard, etc. Assuming we reach such an equilibrium, if more people suddenly want in Dash for just short to medium term commerce uses, then we suddenly need some proportional drop in percentage of Dash stored in MNs (or increase in the speed at which new coins are mined, I guess). If demand suddenly drops from the reached equilibrium, then the percentage stored in MNs need to increase to maintain that optimal price stability, optimal liquidity, and optimal incentives for further proportional investment and growth (and to maintain current investments and growth rate).
Long winded enough?
Honestly, I think the market will regulate this just fine on its own. Unless we see some major liquidity problems, or major growth problems (stalled or recession), or major MN sell-off, I wouldn't worry about it. Just monitor it, find Standard Deviations, establish Confidence Intervals (of say 95% or better), and look for signs of extreme Variance that signal a problem.