A little while ago, I came across a paper (or slides from a talk or something) that seemed to suggest that the modular symbol method for computing elliptic curves over $\mathbf{Q}$ of prescribed conductor $N$ works better when $N$ lacks large prime factors.

I'm trying to find the reference again, but I'm having a really hard time. I have a feeling that it had some connection to John Cremona's program, but this could be wrong.

Could someone please point me in the direction of such a reference?

Did I imagine this paper/presentation? That is to say, is it even true that the modular symbol method is less efficient when $N$ has a large prime factor?

Thanks!

worstcase, based on what I've heard from John about how he does his computations. – David Loeffler Oct 7 '11 at 11:55