Thanks for the advice.
I implemented this approach. Unfortunately not with 100% success.

When enabling sysctl option "net.ipv4.ip_nonlocal_bind = 1" it is possible to use local IP addresses (192.168.1.130-139) as proxy_bind address.
But than using such an address (other than 0.0.0.0), nginx will produce an error message.
Interesting aspect is: attribute "server" in the log entry is empty.
When using 0.0.0.0 as proxy_bind, everything is fine.

Do you have any ideas?

2017/03/09 14:27:09 [crit] 69765#0: *478633 connect() to 192.168.1.21:443 failed (22: Invalid argument) while connecting to upstream, client: x.x.x.x, server: , request: "GET /myservice HTTP/1.1", upstream: "https://192.168.1.21:443/myservice", host: "xxxxxxx:44301"

split_clients "${remote_addr}AAAA" $proxy_ip {
# does not work
100% 192.168.1.130;

# works
100% 0.0.0.0;
}

server {
listen 44301 ssl backlog=163840;
#works
#proxy_bind 0.0.0.0;

#does not work
#proxy_bind 192.168.1.130;

proxy_bind $proxy_ip;