Please note, the SiLabs is not a full-blown ICE; it's what I (for lack of better term - suggest) called here quasi-ICE.
Correct, but not that far from it.
comparing the SILabs JTAG with my Ceibo ICE the glaring difference is that The CEIBO is convenient and bug free, whereas the SILabs is designed to make you excersize your mouse finger as much as possible and have many "small annoying" bugs such as that a change of a bit is lost in the next step.
If SILabs got their act together and fixed the software (the PC part) there would be 2 differnces left: breakpoints limited to 4 and no trace. More breakpoints and trace would, were they there, save me time about 2 times a year.

A general comment: I have worked with ICEs that had trace and, after gaining experience with it, more or less not used it. Finding the cause by poring over 5.000.000 steps of trace takes me longer than breakpointing my way to the goal.

The difference is in the "intrusiveness", or in other words, how close the tool is to the "real stuff" (in each aspect, software, hardware, interfacing, mechanical etc.).
No ICE, quasi or not, should have ANY effect on the operation when it does not stop it. If it affect operations when no breakpoints are activated it is, in my opinion, worthless.

One thing where an ICE is different from "real life" is in the reset, that is a "violation for the sake of convenience" and, are you aware of it, not a problem. As a matter of fact, I have diagnosed a bad reset in 15 seconds by realizing the ICE works, the chip does not.

Erik