That argument is stronger. Your argument was, "If it wasn't there at the start...." which is irrelevant when speaking about proportion.
You substituted "100%" for "majority", which need only mean "50% plus some". Linux was released 23 years ago; ICC version 6 was released in 2002, 12 years ago. That's 11 years before ICC version 6.0 for Linux and 12 years with it; I don't have numbers for pre-6.0, but assume earlier releases came at chronologically earlier points in time. Given its rapid development in that period, the earliest likely release was 2000 or so; but 2002 is the earliest release I have data for.
There have been no other credible compilers for Linux throughout the majority of its existence
Except the Intel C compiler, which is inappropriate for other reasons stated (i.e. it's shitty for non-Intel architectures). Still, given the argument--a GCC bug on x86/x86-64--and the twelve years of potential tuning for icc to support high-performance situations (i.e. embedded architecture, where 16% speed-up matters), broad compiler support is reasonable. It's not like LLVM just becoming useful last year and triggering a scramble to rebase onto CLANG.
2002 was the year of Gnome 2.0, of Linux entering the 2.5 development cycle (2.4 was state-of-the-art), of SuSE 8.0, of single-core CPUs and no AMD64. It was a long time ago, a different age, when journaling file systems were hip and new and Hans Reiser hadn't murdered Nina yet.