Understanding performance of masks: test?a^6 vs. ?a^6test vs. ?a^6

Hey there,

on the same hardware

?a?a?a?a?a?a --- takes ~ 10 Sek at > 70 GH/s
?a?a?a?a?a?atest --- takes ~ 10 Sek at > 70 GH/s
test?a?a?a?a?a?a --- takes about 25 Min at only ~ 500 MH/s

why is it so slow if a prefix is known? It is more common to know a prefix than a suffix, isn't it? - at least that is what I observed.
-S does not improve the performance (it even slows it down quite a lot)

Thanks