A focus of mine is on equal distribution of models across the trading day by session, rather than adding edge/Sharpe without concern for timing vs existing models.
I.e. I would prefer to have 2xAS, 2xLN and 2xNY models at average 0.70 Sh, than I would to have 0xAS, 2xLN and 4xNY models at average 1.50 Sh.
Time is its own form of decorrelation, and suppressed executions from crowding quickly invalidates backtest data on single model performance, no matter how good it appears.
The challenge with Asia in particular is the low volume. I've found that it's easier to find single, high Sh models during LN/NY, but I need to compensate on AS by accumulating very low-freq high Sh models together, in order to balance the trade frequency across sessions.