First, DX10 does *not* in any way perform "better" than DX9- in fact, from an advanced overview, it carries a lot more overhead in ways that could be considered to cause it to perform worse.
What DX10 does have going for it is the *potential* to perform substantially greater for many reasons. The biggest reasons being:
* Moving the hardware abstaction layer (HAL) another step lower and forcing hardware vendors to present/comply to this level. This is a huge step since it will cause greater consistency between different hardware vendors. The way it is now, a single Direct3D capability reported on ATI, NVidia, Intel or S3, when used, can behave four different ways.
The huge benefit here is- developers can push the hardware further with greater assurance that behavior to the end user's will be consistent. This is a HUGE issue now and makes more developers "gun shy" to exploit a particularly neat or cutting edge featureset to it's utmost. This will bring more "console-like" development techniques to the PC since developers will save a TON on QA, testing and user support alone... which in turn makes pushing the envelope much, much more feasible in games.
* Many current limitations and ceilings in the API have been lifted. While not much current hardware can use the full effect of these advances, developers are currently consuming overhead and effort to work-around these limitations, which curbs performance. Again, this will present additional time to tune/improve games as time currently expended for these efforts can be shifted towards engine and content design.
I would not assure any massive performance improvements early-on with DX10 as it will take developers time to get their feet wet and proficient with the new API. Also, realize DX10 will *only* be available for Vista, so the market share of games in development will still favor DX9.0 or backwards paths for DX9.0 in order to make a profit for a while... at least until Vista has a large userbase.
