The original article was very well written and it's sooo nice to statistics to back up the arguments presented.
However, having read the forums here I have to agree with Peso that overall time is perhaps more important than just I/O reads, certainly in terms of customer experience anyway!
I've tried Peso's method on several tables of real data with rows ranging from a few hundred to over 500,000 and in every single case Peso's method is quicker by a factor of at least 10.
Thanks for the post, it has made a massive difference to my paging search times.