A fun endeavor for sure and I like that he measured before & after results for comparison, but I'm not convinced that he made a real improvement. He notes that he saw "a 1db increase in downstream levels occurred along with a full 1db increase in S/N" but 1dB is well within the margin of error and day-to-day variations, especially for the imprecise internal power measurement circuits of the modem. In fact, simply adjusting and re-tightening your coaxial cable can easily result in more than a 1dB change. Also, I'm not quite sure where he came up with the 1000:1 reduction in power rail noise, because his waveforms show significantly less (although still significant) noise reduction. Finally, having a 10-ping average of 7ms vs 6ms is very much within the margin of error one would expect to see.

If he really wanted to improve the RF performance of his modem, he should have used small, low-value, surface-mount capacitors instead of the large tantalum capacitors with relatively long leads and large electrolytic capacitors. Cable modems and other RF equipment operate at very high frequencies by definition, and tantalum & electrolytic capacitors with long leads are ineffective at these high frequencies. The long leads of the capacitors act as inductors, which will prevent the high frequency fluctuations from ever even reaching the capacitor.

If you look closely, you can see that he added his tantalum capacitors to the already-present small ceramic decoupling capacitors. If he really wanted to improve the RF performance of the circuit, he would have been better off adding more low-value, surface-mount ceramic capacitors in parallel with the pre-existing ones to help with the high-frequency decoupling, where it counts in an RF circuit. If he was really adventurous, he could splice some ferrite beads in series with the supply lines to form an LC filter for even better noise rejection.

Furthermore, he could have taken measures to improve the S/N ratio of the circuit by improving the shielding around the sensitive RF circuits in the middle of the board. You can see where the designer originally made room for a shield "can" to be soldered over the sensitive components, indicated by the exposed copper rectangle around the heatsink area. However, you may get unlucky if your shield can volume has resonant frequencies in the operating range of the circuit which will trigger feedback and render the circuit inoperable. A safer option would be to add some RF-absorbing foam over the sensitive area to absorb the noise.