First it seems that the second and third input are reversed. When setting the middle inlet of a [deltaclip~ 0 1] to -1 (to make it work like a [deltaclip~ -1 1]) an incoming signal will only get lower and never grow. Is it that a test is performed on the two values and then sorted internally? That would explain why it would behave like a [deltaclip -1 0]. Reversing the inputs would probably break many existing patches, but the assistance could be changed.
The other issue is that when a clip factor is changed from nonzero to zero, while the incoming signal is negative, the resulting signal very often jumps to zero.
5.0.6 on 10.5.6