A quick analysis of your matrix with 1dsvd shows the following:
## All columns
1dsvd -sing MORLt.xmat.1D'[0..21]'
5078.97 2945.73 1219.04 1182.69 13.5987 10.9621 8.82185 7.50282 6.94958 6.37279 5.71455 5.48894 5.13468 5.03891 4.75331 4.62657 4.39386 4.09438 3.83136 3.74365 3.23214 0.000405762
## Eliminate the last regressor
1dsvd -sing MORLt.xmat.1D'[0..20]'
3008.18 2945.55 1188.49 1182.6 13.5986 10.9621 8.82184 7.50281 6.94958 6.37278 5.71455 5.48893 5.13468 5.0389 4.75331 4.62657 4.39386 4.09438 3.83136 3.74365 3.23213
Removal of the last 'sum' regressor removes the problem, as evidenced by the smallest (last) singular value in each case: a difference of nearly 4 orders of magnitude.
The problem isn't so much the relatively sparse 'silent' times as the fact that the 5th regressor is almost exactly the sum of the first 4 -- it's impossible to tell the combined effects of the first 4 from the 5th.
Your "sanity" check will be the Full F statistic for the 4 regressors, which will tell you if there is any significance in any linear combination of these regressors.
