There is no way to increase a single job's bandwidth greater than a single interface even if you bind 20 nics into one using GEC/LACP etc. If you want more than 1GB/sec with a single backup job, you'll have to go with something like fiberchannel or 10G.
Backing up a single system across gigabit ethernet will never saturate a tape drive. If the tape ever has to pause or stop, the spinup time throws the maximum throughput right out the window. The maximum transfer rate across gigabit ethernet is about 130mb/sec. To achieve that rate, you'd have to backup one huge file. Multiple small files and pausing/waiting for VSS snapshots and other processing tasks limits the bandwidth even further.
One way to get the most out out of tape, schedule multiple jobs concurrently across multiple interfaces to disk storage, then duplicate disk to tape. In this scenario, we are able to achieve more than 1GB/sec and by keeping the tape drive streaming I am seing 8000MB/min on an LTO5 drive.
I can imagine LTO6 would be slightly higher. This aggravated hybrid approach offers faster backups than backing up individual systems directly to tape.
I suppose it would be possible improve on this by backing up half a server like say the C drive (or VM1) on interface x to disk, then duplicate to tape A while simultaneoulsly backing up drive D (or VM2) on interface y to disk then duplicating to tape B. Interesting thoughts