expected/acceptable number of duplicated reads when sequencing human dna, for example at average coverage 1x

Don't think there is an expected/acceptable number. One would always want to minimize PCR/technical duplicates. Some of this is going to depend on quality of the libraries. There are going to applications where duplication would be normal/expected (RNAseq etc).

How to improve the protocol to minimize the duplicated reads?

There are PCR-free methods/kits that minimize duplicates. There will likely be optical/clustering duplicates depending on type of sequencer used (Duplicates on Illumina ).