It appears that what CGEL is saying here is simply that past-form modals in a backshifted context such as past-tense reported speech may be understood as marked for either temporal or modal remoteness, but not both.
[i] He said I could win if I really tried ... may be understood as reflecting EITHER
[a] You can win if you really try ... OR
[b] You could win if you really tried.
[ii] He knew he would be in trouble if they were to check his alibi. ... may be understood as reflecting EITHER
[a] I will be in trouble if they check my alibi ... OR
[b] I would be in trouble if they checked my alibi.
The two [a] readings are what CGEL calls ‘open conditional’; they make a definite but contingent prediction about the outcome. If these readings are adopted, the past forms in both clauses of the embedded conditionals signify a past TO (Reference Time), without ‘modal remoteness’—could and tried match the past tense of said, would and were to match the past tense of knew.
The two [b] readings are what CGEL calls ‘remote l’; they signify that the prediction about the outcome is both contingent and, even if the contingency is met, uncertain. If these readings are adopted, the past forms in both clauses of the embedded conditionals signify ‘modal remoteness’ but say nothing about how TO in the conditionals aligns with the TO in the matrix clause.
CGEL is not saying that there is no Reference Time in the [b] readings of these sentences, merely that the embedded conditionals do not in these readings have independent RTs. They draw their RTs from the clauses in which they are embedded.
It's not clear to me what CGEL is driving at. These observations are true, but as they stand sort of trivial, which from what I know of Profs. Huddleston and Pullum is surprising. I suspect that CGEL employs these observations as a springboard from which to leap into contrasting sentences with embedded conditionals which do exhibit independent RT, either as true irrealis constructions or as inferentials with tense-sensitive conditions and consequences.