Hi @nricks,
@Mehrbod_Estaki raises many great points (thank you!), including how the W values are calculated.
Thus W will depend on the number of features, and you cannot really compare W values between ANCOM tests on two different feature tables. Different threshold W scores will be set for each test.
So what you are seeing is not really a "problem". However, what is a problem (as @Mehrbod_Estaki points out) is that in table #2 many many features are significantly different, which is breaking the assumptions of ANCOM.
You might want to check out q2-gneiss; especially since it sounds like you have a potentially complicated experimental setup, gneiss can handle things like multi-factorial experimental designs to determine how species balances differ between groups.
I hope that helps!