As far as I know the effectiveness of chunking depends on (1) test speed and (2) amount of practice.
At low speed simple techniquest benefit almost everyone a little, even with virtually no practice. Quoting from Dempster and Zinkgraf
Recently, in fact, evidence has been found that questions the validity of the assumption
that chunking plays a role in digit-span variance. In one experiment,
Lyon (1977) presented 12-item digit sequences to adult subjects for whom conventional
digit-span estimates were available, and instructed them to think of each
successive group of three digits as a three-digit number. The important result was
that subjects with high spans (average = 8.3) benefited just as much from the chunking
instructions as those with low spans (average = 6.1). Thus, prompting all subjects to
impose a common structure on the lists did not produce the convergence of the two
groups that would be expected if chunking was the source of differences between
subjects in the high- and low-span groups.
On the other hand
However, it does seem plausible that the temporal characteristics of
the digit-span task (rapid presentation rate and single trial) simply allow little opportunity
for chunking, unless the individual is exceptionally adept at creating
chunks (Hunt & Love, 1972, pp. 248-250), or has received an extended period of
practice (Ericsson, Chase, & Faloon, 1980; Gates & Taylor, 1925; Martin &
Fernberger, 1929). For such an individual, chunking can be a highly successful
strategy.
The "extended practice" mentioned there involves associative techniques not merely chunking.