There’s more than one way to calculate RMS (1) and I think ACX keeps their details secret so it’s probably best to trust Audio Lab. If ACX check is too close to the limits you might fail.
The peak readings are simpler and they should match, but there is also more than one way to measure noise.
Boosting by 1.4dB should increase the Audio Lab reading to -22.7dB and I assume that will pass. But it will make your peaks very close to the -3dB limit.
Or, the Audiobook Mastering Macro has some “wiggle room”, putting you in the middle of the RMS spec and leaving some headroom in the peaks so it should work every time.
(1) I believe it’s impossible, or at lease impractical, to calculate RMS for the whole file so it’s time-windowed and it’s an average of RMS calculations or an RMS of RMS.