evalv3: repeated use of schema causes wall time and memory use to blow up

[mvdan]

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c

-- input.cue --
package p

import "list"

#Schema: {
	required?: [...string] 
	properties?: [string]: null | bool | string | [...int] | #Schema

	foo00?: null | bool | string | [...int] | {string: int32}
	foo01?: null | bool | string | [...int] | {string: int32}
	foo02?: null | bool | string | [...int] | {string: int32}
	foo03?: null | bool | string | [...int] | {string: int32}
	foo04?: null | bool | string | [...int] | {string: int32}
	foo05?: null | bool | string | [...int] | {string: int32}
	foo06?: null | bool | string | [...int] | {string: int32}
	foo07?: null | bool | string | [...int] | {string: int32}
	foo08?: null | bool | string | [...int] | {string: int32}
	foo09?: null | bool | string | [...int] | {string: int32}
	foo10?: null | bool | string | [...int] | {string: int32}
	foo11?: null | bool | string | [...int] | {string: int32}
	foo12?: null | bool | string | [...int] | {string: int32}
	foo13?: null | bool | string | [...int] | {string: int32}
	foo14?: null | bool | string | [...int] | {string: int32}
	foo15?: null | bool | string | [...int] | {string: int32}
	foo16?: null | bool | string | [...int] | {string: int32}
	foo17?: null | bool | string | [...int] | {string: int32}
	foo18?: null | bool | string | [...int] | {string: int32}
	foo19?: null | bool | string | [...int] | {string: int32}
}

out: [...#Schema] & list.Repeat([{
	#Schema
	properties: foo: required: ["bar", "baz"]
}], 100)

As of f2c38f8:

# With the old evaluator. (0.411s)
> env CUE_EXPERIMENT=evalv3=0
> exec cue vet -c
# With the new evaluator. (4.120s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue vet -c

The peak memory usage also jumps from 63MiB to 3.4GiB, meaning we allocate nearly one gigabyte per second. So something seems very wrong.