It appears the juce::dsp::ConvolutionEngineFactory constructor sets the lower limit for an FFT based convolution arbitrarily to 64. So for a juce::dsp::Convolution, even if juce::dsp::Convolution::Latency is set to a smaller power of 2, and the process spec max block size is also a smaller power of two, the FFT (and thus latency) will still be 64. Is there some reason this lower limit isn’t 32, 16, 8, etc.?