Not sure: as an experiment:
Start: ent0, ent1 (to become link aggregate), ent2 backup adapter.
1. Create link aggregate using ent0 and ent1 as physical interfaces -> creates ent3
2. Using NIB use ent3 as main interface and ent2 as backup to create NIB interface ent4
3. Test using ent4 as interface.

I may be way off on this, but maybe this will give you what you are looking for. I lack the hardware to test this myself.

Well 3Gb would be 384MB and 1Gb 128MB.
So I would think your monitored 95MB throughput is more like 1Gb than 3Gb.

Have you selected Round Robin mode of operation for this EtherChannel ?
(Backup mode will only use the backup adapters if the main ones fail, and Standard mode may be the cause of the lack of monitorable performance.)

Is the switch to which this connects definitely capable of the Cisco EtherChannel linakage ?

Incidentally, I've adapted an earlier posting on this topic for the articles section, which Michael is doing some better screen shots for, and then it should be easier to find and have some focus.  We think Link Aggregation (the improved open version of EtherChannel) is really great.  The recommendation we make for anyone with more than one ethernet NIC port, is to use at least the Backup mode of the Link Aggregation instead of the EtherChannel as such - that has no requirements on switch config while giving you the resilience like an HACMP monitored LAN link.

However, obviously you want the enhanced throughput of either the standard or round robin modes to be using all the adapters simultaneously.  Now if you use the standard mode, a conversation will only be taking place up the adapter it's started on for the duration, whereas the round robin mode will move the traffic over all adapters in the link.  So what you're seeing there might be because you have used standard mode when I think you wanted round robin.