It is possible that to improve system performance, you could more your nonclustered indexes to other file groups and have this file groups on separate spindles. This could allow for parallel access between the NCI and the CI possibly improving performance for bookmark lookups. Also, if the NCI are covering indexes and are on separate spindle(s), this would reduce reads from the file group and disks where the CI reside.

All of this could help increase performance. YMMV, so this needs to be tested completely in a test environment before being implemented in production.