The extraction tool fractures the metal polygons into pieces, and assigns resistance values to each of them, based on the fracture length, width, and sheet rho of the metal layer. This is 1D R extraction. (QRC can also do 2D R extraction, but it's not the baseline algorithm used for R extraction).
If metal layout is more or less one dimensional (40 squares is pretty much 1D), this approach should work well, in a sense it should be giving accurate resistance value (say within a few percents).
However, if individual fracture resistance is less than a critical/threshold value (I think, typically 0.001 Ohm - i.e. 1 mOhm) - these resistance may be "reduced" by simply shorting (ignoring them). If you have many less than 1 mOhm resistors in series - here is one possible root cause for R accuracy loss.
Second (after fixing the first issue described above), you can look at the post-layout netlist (you need to generate DSPF file - which is a straightforward text file format - rather than extracted view or calibre view), and look at individual resistances, their L, W, and R values - and check if that is correct. Calibre XRC should have a pretty good facilities for visualizing extracted resistors. QRC's one is not that good (in extracted view, you can hardly see parasitic resistors).
What's the sheet resistance of the metal layer that you have problem with?
In general, if layout shapes are not simple 1D polygons, R extraction as done by most of the standard extraction tools is not very accurate, especially at low resistance values - typical examples are power transistors with trapezoidal or other complex shapes in upper metal layers.