Can we explain why Timing is inconsistent?

Question

I am trying to speed up a code and by placing Timing statements in the code I identified a suspect calculation that I do repeatedly. In order to test out the calculation I generated some random numbers and looked at the timing of the calculation outside the main code. Here is an example of doing the calculation a number of times.

a = Table[
  n1 = 10000;
  n2 = 50;
  cmatT = RandomComplex[{-1. + -I, 1. + I}, {n2, n1}];
  w = RandomComplex[{-1 + -I, 1 + I}, n2]; 
  Timing[Transpose[cmatT] . w;][[1]], {50}]

The output is

{0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., \
0.015625, 0., 0., 0., 0., 0., 0., 0.015625, 0.015625, 0., 0., 0., 0., \
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., \
0., 0., 0., 0., 0.}

So the timing is inconsistent with three cases taking a long time and the remainder being very fast. In my code it always takes a long time. If you run the above code repeatedly then the long times appear inconsistently in the output.

To try to understand the output I stored the matrices involved and ran the following code

a = Table[n1 = 10000;
   n2 = 50;
   cmatT = RandomComplex[{-1. + -I, 1. + I}, {n2, n1}];
   w = RandomComplex[{-1 + -I, 1 + I}, n2];
   {cmatT, w, Timing[Transpose[cmatT] . w;][[1]]}, {50}];

I tried to find an example where the time was significant and see where the problem lies using the following

{tim} = Select[a[[All, 3]], # != 0. &, 1];
n = Position[a[[All, 3]], tim][[1, 1]];
cmatT = a[[n, 1]]; w = a[[n, 2]];
Timing[Transpose[cmatT] . w;][[1]]

Output

0.

So what is happening? I expected one of the same times as I had previously. I suspect this is something more to do with the way Timing works rather than my code. I have tried to run ClearSystemCache[] to see if this was an issue without success.

Any ideas? What is the correct way to work out this sort of problem?

Edit

Following the suggestion of MarcoB I tried AbsoluteTiming instead of Timing. The code is now

a = Table[n1 = 10000;
  n2 = 50;
  cmatT = RandomComplex[{-1. + -I, 1. + I}, {n2, n1}];
  w = RandomComplex[{-1 + -I, 1 + I}, n2];
  AbsoluteTiming[Transpose[cmatT] . w;][[1]], {50}]

with output

{0.0020004, 0.0017151, 0.0033361, 0.002125, 0.0017301, 0.0017893, \
0.0015143, 0.0014935, 0.0015266, 0.0015972, 0.0017455, 0.0014958, \
0.0014959, 0.0014849, 0.0014875, 0.0014839, 0.0014996, 0.0015271, \
0.0015269, 0.0014784, 0.0016972, 0.0015954, 0.0015108, 0.0016923, \
0.0015294, 0.0015364, 0.0017162, 0.0015108, 0.0017251, 0.0016942, \
0.0017495, 0.0017094, 0.0015394, 0.0016843, 0.0015176, 0.0015188, \
0.0015184, 0.0015009, 0.0015237, 0.0014876, 0.0015088, 0.0015569, \
0.0016929, 0.0015513, 0.0016194, 0.0015381, 0.0016187, 0.0014727, \
0.0020828, 0.001443}

which makes much more sense.

Can someone check to see if this is a bug? My Version is 13.3 Windows