Results for years 4–8 of a long‐term grazing experiment on swards of a diploid perennial ryegrass (Lolium perenne), var. Contender (D swards), a tetraploid ryegrass, var. Condesa (T swards) and Condesa with S184 white clover (Trifolium repens) (TC swards), direct sown in May 1987, are presented. The swards were continuously stocked with sheep from 1988 to 1990, as previously reported, and for a further 5 years, 1991–95, at a target sward surface height (SSH) of 4–6 cm. Control of sward height was successfully achieved by variable stocking, except in 1993 when paddocks were set stocked and the resulting mean SSH was 9·3 cm. Grass swards received on average 160 kg N ha−1 year−1; grass/clover swards were mainly not fertilized with N with the exception that they were given 30 kg N ha−1 as a remedial mid‐summer application during a period of low herbage mass on offer in 1994 and 1995.Mean white clover content of the swards fell from 18·2% of herbage dry‐matter (DM) in 1992 to 8·5% in 1993, whereas stolon lengths fell from 120 to 58 m m−2. A return to lower sward heights in 1994–95 resulted in an increase in white clover content to 12·8% by the final sampling in August 1995. Perennial ryegrass content of the grass swards remained high throughout (mean 96·7% in 1995). Perennial ryegrass tiller densities recorded in August 1991, 1993 and 1994 showed consistently significant (P < 0·001) sward differences (3‐year mean 16 600, 13 700 and 10 100 perennial ryegrass tillers m−2 for the D, T and TC swards). In 1994, the year after lax grazing, a low perennial ryegrass tiller density (9100 m−2) and low white clover content (mean 4·3%) in the TC swards resulted in a much lower herbage bulk density than in the grass swards (April–July means 72, 94 and 44 kg OM ha−1 cm−1 for the D, T and TC swards). There was a consistent 40 g d−1 increase in lamb liveweight gain on the TC swards over the T swards, except in 1994. In that year there was a reduction in lamb liveweight gain of 33 g d−1 on the TC swards and a significant increase in ewe liveweight loss (117 g d−1) associated with low herbage bulk density despite optimal sward height. Lamb output (kg liveweight ha−1) on TC swards reflected white clover content, falling from a similar output to that produced from grass given 160 kg N ha−1, at 18% white clover DM content, down to 60% of grass + N swards with around 5% clover. A 6% greater output from the T than the D swards was achieved mainly through higher stocking rate. The experiment demonstrated a rapid, loss in white clover under lax grazing, and showed that the relationship between performance and sward height is also dependent on herbage density. High lamb output from a grass/clover sward was only achieved when the clover content was maintained at 15–20% of the herbage DM.