The most successful schools appear to be increasing the proportion of pupils gaining at least five A*-C GCSE grades at three times the rate of their most disadvantaged rivals.
An analysis of postcodes has confirmed that high-performing schools are attracting a growing proportion of socially-advantaged pupils.
Some academics have recently argued that Conservative education reforms have not been as socially divisive as was predicted. One University of Wales study showed that free school meal entitlement - a key indicator of child poverty - is more evenly spread in England and Wales than it was in 1988.
But the new study by Alex Gibson of Exeter University and Sheena Asthana, of Plymouth University, suggests that the research into free meal entitlement was "simplistic".
Gibson and Asthana, who presented their findings to last week's conference of the American Educational Research Association, say they have proof that the gulf between the strongest and weakest schools has been growing steadily since the early 1990s.
They analysed the GCSE results for 1,584 comprehensives and found that 71 per cent of pupils in the top 10 per cent of schools gained at least five GCSE A*-C grades on average last year, compared with 65 per cent in 1994.
The corresponding figures for the bottom 10 per cent of schools were 13.1 per cent in 1998 and 10.6 per cent in 1994.
"The difference between the average GCSE performance of the top and bottom 10 per cent of schools therefore increased from 54.5 per cent to 57.9 per cent between 1994 and 1998," the researchers say.
But they add that the key question is whether a school's position in the local pecking order has influenced its rate of improvement or the social composition of its pupil roll.
They carried out a second study of 457 schools which examined not only GCSE results but pupils' home postcodes. They then compared each school with four others with overlapping catchment areas.
"This revealed that the percentage of pupils achieving five or more GCSE A*-Cs in the 75 top-ranked schools has increased by, on average, no less than 1.65 per cent per annum, whereas schools ranked lowest among their local market groups have only increased this key rate by 0.57 per cent each year," they say.
"This is a very substantial differential and will, very rapidly, have a noticeable effect ... the higher a school's local rank the more substantial has been its reduction in the percentage of pupils in receipt of free school meals.
"Within local markets, the evidence is clear that high-performing schools both improve their GCSE performance fastest and draw to themselves the most socially-advantaged pupils."
AERA report, 16