Lots of excellent info from c6em :-)
There is a very good paper on xDSL. It was authored by Matthew Ernelli, an employee at Telenor, Norway's incumbent telco. [1]
In chapter 3 of his report, Ernelli performs lab tests using two different CPE modems (Broadcom & Texas chipsets) and two different DSLAMs (Conexant & Broadcom chipsets).
In the lab set up, he forms a 0-loop (zero length loop) between the Central Office kit (the DSLAM) and the Remote Equipment (the CPE modem). The 0-loop was kept as short as possible, so as to minimise the effect the cable might have on the measurements.
The measurements obtained from each of the modems (QLN and Hlin/Hlog) all had to be normalised before comparisons could be performed.
Ernelli offers no specific conclusions but it's clear that he wasn't impressed with the Texas chipset.
This would make an excellent project. Probably the DSLAM kit is just too expensive to do it for anything other than a commercial project. No doubt big telcos like British Telecom have performed their own extensive in-house testing of all major chipsets.
I'm no authority at all, but would guess that the most critical component is the digital signal processor and the competence of the DSP codecs at recovering the symbols from the QAM scatterplot. Different codecs will perform this better in different conditions, e.g. under varying levels of noise. There must be many other issues though. The sensitivity of the receiver is doubtless important. Maybe it is complicated by non-linearity of the signal amps in the modem. Internal noise seemed to be a problem for the (unidentified) chipset from Texas.
cheers, a
[1]
https://eeweb01.ee.kth.se/upload/publications/reports/2008/XR-EE-KT_2008_003.pdf