CIR Estimation: MPCs Amplitude Differences (FFT vs. xcorr)?

When estimating the channel impulse response (CIR) using FFT-based correlation vs. xcorr() in MATLAB, I notice that while the main peak aligns well, the amplitudes of MPCs differ.

What could explain this amplitude mismatch? Which method should be trusted more for accurate CIR amplitude estimation: xcorr() or FFT-based?

Thank you!

[dpb]

You neglected to show us how you calculated either way, but both can be normalized in many ways; how, specifically will depend upon what the actual definifion of the computed value uses.