This issue with the apparent idle sound and on-load vs. off-load sounds comes back to the same old problem: a lack of a dedicated intake source, including originally in recordings made (part throttle only). That is the key differentiator of "load" - i.e. throttle position and its effect on the sound that escapes past it.
By mixing the intake in with the exhaust, it also creates the issue of overly similar sounds in different views, and from different angles around the cars, as though they've lost their confidence in accurately mixing the sources they do use. I say this because without intake (and other sources), it sounds thin and incomplete - i.e. the GT of the past that "everyone" complained about.
This newfound, but fixed, intake contribution explains the good interior mix, where intake sounds are most prominent in almost every car, but that unfortunately does not translate to both an expressive (vs. inputs) and accurate external sound. The samples in use for the exhaust source are actually often intake heavy, yielding a result much like the intake-heavy mixing of the Forza series.
GT5/6 was much, much (combined with reverb, delay and low-passing: almost infinitely) better in terms of external mixing, except that it still had mostly no intake sound whatsoever, and that utter lack of intake sound meant interiors were especially lacklustre as a result.
AES only creates exhaust sounds in its current form, and with that PD have made significant quality improvements in GT Sport, but because they toned down its character (distortion effects) and the volume level it's not so noticeable. They also blend it with samples to good effect across most cars (they played samples at the same time on PS3, but they were always almost inaudibly quiet and not really "matched" to the synthesiser output properly in any way).
GT6 already had the framework in place for an intake source with AES, they are just yet to use it for reasons unknown. I personally found it harder to synthesise intake sounds, but I recently discovered* a little trick that does a reasonable job. The problem is that I did it brute-force style, whereas Sony have developed a seemingly more elegant, certainly more efficient solution (because of the multiple sources in use), and it will require a substantial amount of tedious maths-ing to implement "my" trick into their system, or any other similar method.
Once they do that, and especially if they revert to their old mixing method with it, AES will be so expressive that the remaining small differences in absolute quality / fidelity, relative to mashing recordings together, will not be an issue.
* I actually tried it several years ago, but clearly didn't implement it properly. I almost did it wrong again this last time...
