Terrence,

I think you are better off preprocessing (=transforming) the incoming text so that you can mark the end of the second capture, so that things look like:
%<ufm>AAA<ufn>[tag]<index>B1</index> The following tables summarize..

Like this you do not have to capture the whole para in xyPerl. You do not want to do this because it makes editing the text in the xyview a lot more difficult.

As for the <index>..</index> macros, you define them to call a xyPerl function with the 'i' mode flag set so that you capture just the B1 thing.
remark 1: do not forget to sue the '+' sign as the first character in the expansion of the <index> macro.
remark 2: do not forget to strip the <index> and </index> macros from the captured text in your xyPerl

As for storing multiple occurrences of the <index> thing on the same page, there are several solutions possible.
Probably the easiest one is to use an array (@Index) and push each capture onto the array.
And then during frills processing you unshift each index term every time you hit a line with an <index> macro.
(you will need to flag these line and find a way of knowing if the flag is used for a margin note or an index term).
At the end of the frills processing you undef the @index arrary so it is ready for the next page.

You might come up with another solution that suits your needs better.
Like if the index term always appears where there is a margin note, you might put both pieces of text into the 0 register using a slightly different transformation and split them later using xyPerl.

Good luck.

Bart