It's not bullshit, it's a fact of advanced FFT design that anyone who has made a FFT with "spectrum-analyzer-like controls" will be able to tell you.
Averaging M FFTs of N points is is mathematically equivalent to calculating the FFT of the M*N points and taking every Mth point of the spectrum, but faster -- O(n) instead of O(n*ln(n) in time and O(1) instead of O(n) in memory. Multiply the M*N time-domain samples by a window function before taking the FFT and you convolve/smooth the frequency response that you would get *before* downsampling -- that is, you mix information from the bins that the fast technique discards into the bins it doesn't discard, resulting in the same process gain / noise floor and scalloping improvements as the M*N FFT while maintaining O(n) time cost and O(1) memory cost. Obviously with O(1) memory cost you can keep averaging until the cows come home, so it doesn't really make much sense to say it has a certain size to it.
If you actually want M*N frequency bins, you're out of luck and you have to pay full t=O(n*ln(n)) s=O(n) price. If you just want the noise floor & scalloping benefits of a M*N FFT, you only have to pay t=O(n) s=O(1).
> charging extra for something like 16bit HiRes
Now *that* is bullshit.