Boot vs. MTL combined with ITL

Hello,

I'm a user of the MTL library, and I use it in conjunction with ITL. 
(Specifically, I make heavy use of bicgstab with SSOR and modified
cholkesky preconditioners.)

Recently, I have read about a few people switching over to using
Boost/uBLAS.  I'm interested in hearing about anyone's experiences in
switching their code from MTL to uBLAS, and am specifically wondering:

(1) Can you comment on any performance improvement (or decrease) in
general that you noticed after switching your matrix package to uBLAS
from MTL?

(2) Most importantly, how well do the ITL solvers and preconditioners
work with uBLAS sparse matrices?  Based on my reading, if I understand
correctly, ITL preconditioners don't work entirely well with uBLAS, but
what I read could have been outdated. 

Right now, using MTL and ITL (with sparse matrices) works fine, but I'm
always open to performance improvement, especially if it's as simple as
switching the underlying matrix routines :) ... 

Any thoughts on this subject would be welcome.

Thanks, 
Kris

_______________________________________________
This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/