On a wavefrontier T90 in Belgium, I get:
12639H: 9dB FEC 3/5
12643V: 10.1dB FEC 2/3
11636V: 11.9dB FEC 2/3
So slightly weaker signal indeed on 12639H and because of the more critical FEC 3/5 a higher SNR is needed
for correct reception. 9dB is enough here in Belgium (5.5dB would be the absolute minimum needed)
On smaller dishes there could also be some influence from neighboring satellites (19.2 E 12633H)
especially if the dish would be misaligned. A problem with the skew could also explain it.
Below is a spectrum showing the overlap on a correcly skewed lnb. Note that the indicated frequencies and synbolrates
are not the true ones, but the ones which are good candidates to start a blindscan.