Maintain 16-bit throughout the HDR process

I've been doing some HDR stuff lately and I've been plagued by image noise. I've read that HDR photos using only 3 exposures are prone to noise, especially in darker areas of the image, and my Rebel XTi only does 3-exposure bracketing (and I don't feel like venturing into trying to do more by hand).

Anyways, this is my workflow:
I take 3 images in RAW, 2 stops apart.
I bring them into Adobe Photoshop CS2 and do Automate>Merge to HDR.
I save it in the OpenEXR format. This reduces the file size to around 28 megabytes.
I open the .exr file in Qtpfsgui, tonemap it using the default settings for the Manituk algorithm (sometimes boosting the saturation).
Then I save that as a .bmp and bring it back into Photoshop and convert it to 16-bit.
I'll usually take the middle exposure, overlay it over the tonemapped .bmp, and play with the opacity until it looks good (and then I'll usually do a few other things to the contrast and saturation).

So the problem, I think, is when I save it and take it to Qtpfsgui. The output from Photoshop's 'Merge to HDR' tool is a 32-bit photo, but when I save it as OpenEXR, 28MB for a 10 megapixel photo doesn't sound like it'll be 32 bits. More like 8. Then when I save the tone-mapped one, it's saved as a .bmp which is also about 28MB. That sounds like another 8-bit photo.

Does anyone use Photoshop and Qtpfsgui for HDR photos? Can Qtpfsgui import any 32-bit image formats that Photoshop CS2 exports? I know that Qtpfsgui also merges photos to HDR but I like the results from Photoshop alot better. I don't want to go 32-bit -> 8-bit -> 8-bit -> 16-bit -> 8-bit. I'd rather go 32-bit-> 16-bit -> 8-bit.