[VC2010] Suggestion for higher perf debug builds

[MaxDZ8]

Although profiling debug builds makes little sense, I want them to be at least more or less realtime.
This is not the case for the game I'm working on. It has not been the case for a while but I am now quite worried.
I suppose I might be doing something wrong and I'd like to get some suggestions on how to improve Bullet's performance while in debug mode.
I'm currently hitting about 103 ms/frame, with btDbvtBroadphase, bt32BitAxisSweep3, btAxisSweep3 all performing more or less equal.

I have run the game through CodeAnalyst: on a total of 372900 CPU clocks,
57792 (16%) are spent in btCollisionDispatcher::defaultNearCallback
12976 (3%) are spent in btCollisionObject::isActive
12714 (3%) are spent in btCollisionObject::getActivationState

The first function I recognize as mine takes 0.10% of total samples, there are still bullet calls beyond this point making bullet taking about 92% of frame time (which includes an highly inefficient Render() call in D3D9).

I suppose a release build would inline a lot of things, making my concerns go away, but I wonder how bigger products are tested in debug when using this.