[Performance] consolidate TDs in ParallelEnv without buffers #10470

Job	Run time
build-wheel-linux (3.8, cp38-cp38, --index-url https://download.pytorch.org/whl/cpu, "['cpu', '11...	1m 32s
build-wheel-windows (3.8, 3.8)	2m 33s
build-wheel-windows (3.9, 3.9)	2m 22s
build-wheel-windows (3.10, 3.10.3)	3m 35s
build-wheel-linux (3.9, cp39-cp39, --index-url https://download.pytorch.org/whl/cpu, "['cpu', '11...	1m 34s
build-wheel-windows (3.11, 3.11)	2m 30s
build-wheel-linux (3.10, cp310-cp310, --index-url https://download.pytorch.org/whl/cpu, "['cpu', ...	1m 39s
build-wheel-linux (3.11, cp311-cp311, --index-url https://download.pytorch.org/whl/cpu, "['cpu', ...	1m 41s
test-wheel (linux, ubuntu-20.04, 3.8)	2m 13s
test-wheel (linux, ubuntu-20.04, 3.9)	2m 20s
test-wheel (linux, ubuntu-20.04, 3.10)	2m 17s
test-wheel (linux, ubuntu-20.04, 3.11)	2m 31s
test-wheel-windows (3.8)	4m 42s
test-wheel-windows (3.9)	3m 39s
test-wheel-windows (3.10)	4m 52s
test-wheel-windows (3.11)	5m 3s
	45m 3s

Provide feedback