Actually, not all downconverting is created equal. I find serious audible problems to occur when downsampling/downconverting in certain software. Godlwave, for example, only caused audible problems when attempting to use it's automated downsample/downconversion options.
Cool Edit Pro 2, I found to be excellent(transparent in ABX tests compared to the original 96/24 hi res files) when first prefiltering with a high quality FFT filter to make sure no spectra over 22kHz is left. Downsampling uisng the highest quality mode. Then using a high quality 1.2 bit triangular dither to reduce to 16bit space. BTW, Dither noise must be done when A-D sampling or when reducing a word space in software. Failing to do this results in quantization error, that can become an audible problem in 16 bit space.
As you said, it is simpler to just sample at 44.1/16 to begin with, assuming one has a card that natively samples at that rate with high quality.
-Chris