Around four years ago I ran some extensive tests of AIM over a range of periods and assets and came to the conclusion that if you weighted to the same average stock/cash weightings as what AIM averaged over the investment period and periodically rebalanced (yearly or at bands such as +/- 40% i.e. start with 25% weight and rebalance back to 25% target weighting if the weighting declined to 15% or less or rose to 35% or more) then you came out with similar overall results. In that context AIM might lose out in practice as it tends to trade more often (higher trading costs). The plus side however is that AIM automatically navigated to reasonable weightings - whereas that average weighting couldn't have been known in advance for the fixed weighting approach.
If for instance it was more appropriate to initially weight to say 40% stocks/60% cash fixed weightings over the investment period than it was to weight to 50-50, then AIM would tend to have averaged closer to 40-60 weightings over the investment period.
MPT efficient frontier is a conceptual thing. The efficient frontier is dynamic and can't be predicted in advance, only measured with hindsight. AIM has the tendency to navigate close to the efficient frontier. Whilst (generally) there will be more (and less) rewarding alternatives, being close to the efficient frontier provides the better risk adjusted reward.