I believe the approach you are taking moving openal into direct sound is not a good idea and is going to create a lot of regressions that can not be fixed with your current approach.
Windows audio programmers know that they can get access to the same hardware through multiple APIs at the same time. That's because all the APIs eventually go through the same driver to access the hardware.
A common scenario for voip type applications using direct sound is to find the corresponding wave device using the private property set and then find the mixer device from the wave device so they can get access to the mixer volume to implement microphone AGC.
The right approach for using openal is to develop an openal driver like our current oss and alsa drivers. That way all APIs will have the same consistent view of the hardware. The direct sound driver API already supports multiple hardware buffer access so the software mixer can be bypassed.
You need to look at voip and audio conferencing applications. You also need to look at XP era utilities and games that came with boxed Creative sound cards. You could also try XP era openal programs using the supplied openal dll which is just a thin layer over direct sound.