polcompositum() vs factornf() performance comparison

[Igor Schein <igor@txc.com>]

Hi,

? p1= x^16+8*x^14+148*x^12-184*x^10+5680*x^8-102304*x^6+546088*x^4-1557232*x^2+4351396;
? p2=x^16+8*x^14-152*x^12+416*x^10+13030*x^8-12904*x^6+15688*x^4+954368*x^2+657721;
? #
   timer = 1 (on)
? poldegree((p=polcompositum(p1,p2))[#p])
time = 1,110 ms.
32
? matsize(factornf(p1,subst(p2,x,y)))
time = 2,260 ms.
[8, 2]

Basically, I am verifying in 2 different ways that 2 number fields are
non-isomorphic, but it turns out factornf ( for which nfisisom is a
wrapper ) is twice slower!!  

Any comment?

Thanks

Igor