I am assuming you're using "bridge-mode" for networking (your internal, virtual adapter is bridged to your host's physical adapter).
In any case (unless you explicitlly manually set them to the same address, which causes a lot of other problems), your guest (WinXP) machine will have a different MAC address than your host (CentOS). Due to bridge-mode, your host acts as an ethernet switch, and forwards packets to the guest.
So when an ARP broadcast comes, your host gets the packet and also forwards it to your guest machine. When packets for your host come, they are processed by your host's network stack. When packets for your guest come, the host forwards them to your guest, which then processes them as needed.
I'm making some assumptions on your setup and how exactly your ISP is giving you these IPs, so if any of this is wrong I apologize and will happily change my answer
For your internal network I would suggest you setup a DHCP pool for your workstations and statically assign IPs to your servers. I'll leave the DHCP pool setup for you, as I think you're mainly aiming to make sure both public IPs are utilized by the proper networks.
i.e.
172.16.1.0/24 for your workstations, with DHCP, assigned to VLAN10
172.16.2.0/29 for your servers, statically assigned, on VLAN20
That all being said here is what I personally would try and setup to get your gear online.
int g0/0
ip address dhcp
This will pull an IP from your modem and give it to your external port. I suspect it will be an ISP internal IP because I doubt they'd give your modem a publicly routable IP. That'd be weird.
In this scenario, you should not be manually inputting any default routes on your router as it should all be supplied from the DHCP pull.
int g0/1.10
ip address 172.16.1.1 255.255.255.0
int g0/1.20
ip address 172.16.2.1 255.255.255.248
This setups the internal gateways for your two networks. So all your workstations will be pointing to 172.16.1.1 and your servers to 172.16.2.1
After that you'll need to setup NAT rules on the router to handle passing of traffic outwards for your workstations.
int g0/0
ip nat outside
This setups your external facing interface as your outside nat interface.
int g0/1.10
ip nat inside
This setups your internal facing interface as an inside nat interface.
Router(config)# ip nat pool internet 128.66.0.2 128.66.0.2 prefix 24
Creates a NAT pool named internet being translated to one of your public IPs.
Router(config)# ip nat inside source list 7 pool name internet overload
This says to NAT all IPs in list 7 to the NAT pool you just created and that you can overload it. Which is to say more than one internal IP can use the same external IP.
Router(config)# access-list 7 172.16.1.0 0.0.0.255
Creates the list referenced in the previous command. Now onto NAT for your servers, which I suggest be statically assigned if you want them publicly available.
int g0/1.20
ip nat inside
Same as before, this setups your internal interface as an inside NAT interface.
Router(config)# ip nat inside source static 172.16.2.(2-6) 128.66.1.(2-6)
A new line for each static assignment is needed. This creates a static translation between your internal IP and your external IP that was assigned to you.
As for your switch; all you would need to do is properly tag your ports depending on what is plugged in and make sure your trunk is passing both VLANs.
At this point both subnets should hitting your router, and your router should know where to pass the traffic, be it internally (your workstations getting to your servers) or externally (internet). Access control can either be setup with ACLs on the router, a stand-alone firewall, or firewalls on your servers.
Now this all hinges on how your ISP has your modem setup. If it works the way I think it works, when your external interface pulls it's information through DHCP, your router should populate both your public IP ranges so that when your router NATs it knows where to send your traffic.
I suspect someone will give a better written answer, but hopefully this points you in the correct direction.
I also referenced the following link for help on the NAT parts as they are definitely not something I play with very often.
http://www.cisco.com/c/en/us/support/docs/ip/network-address-translation-nat/13772-12.html
Best Answer
Let's start out with the remark that routers not doing proxy arp is the default situation, proxy arp is disabled in most networks, and, if not used carefully, it can break a lot of things in nasty ways.
In the normal situation with proxy ARP disabled your client will see that the other machine is not on its local network. So it looks at its routing table and see that it should reach the other network via the router. Then, it will forward packets there.
As for your second question: one example of that is a misconfiguration in netmasks. A machine with 10.0.0.1/24 can think it's netmask was configured as a /16 because it sees ARP entries for 10.0.1.1/16, while the latter is not on the local network.