Efficient PCR with 3' mismatch?

[David R. Johnson]

Dear netters,
   After RT-PCR amplification of a highly polymorphic gene family (HLA), 
we have observed reasonable yields of a product that is mismatched from 
the primer at the 3' end nucleotide (even one with the last 3 
nucleotides mismatching).  The remainder of the primer matches 
perfectly.  Is this a common occurance?  Can anyone suggest ways to 
reduce or avoid this problem?  

PCR conditions are:
35 cycles of
95oC 30s denature
62oC 30s anneal (20 nt primers, Tm ~62oC)
72oC 30s extend (~1kb product)
template = oligo dT primed, Superscript reverse transcribed cDNA

thanks for any suggestions,
Dave